tag:blogger.com,1999:blog-6141980.post4645994940669354317..comments2024-03-20T12:28:35.004-05:00Comments on Nuit Blanche: CS: Remarks on Bob's exploration of the failings of OMP and l_1 MinimizationIgorhttp://www.blogger.com/profile/17474880327699002140noreply@blogger.comBlogger4125tag:blogger.com,1999:blog-6141980.post-42909281435170607862010-12-13T15:07:33.834-06:002010-12-13T15:07:33.834-06:00Alejandro,
No. You are probably not missing anyth...Alejandro,<br /><br />No. You are probably not missing anything. Thanks for the information. <br /><br />Cheers,<br /><br />Igor.Igorhttps://www.blogger.com/profile/17474880327699002140noreply@blogger.comtag:blogger.com,1999:blog-6141980.post-89293027771986781482010-12-13T14:07:33.091-06:002010-12-13T14:07:33.091-06:00But, as far as the solution to the optimization pr...But, as far as the solution to the optimization problem is unique, shouldn't all the solver gives you the same solution up to the numerical tolerance? <br /><br />I've tried solving a BP problem with CVX and l1_magic, and they produce essentially the same results. And if you look at Bob's experiments, the first thing he tried was to replace CVX by the linprog solver, getting the same results.<br /><br />Am I missing something here?Alejandro Weinsteinnoreply@blogger.comtag:blogger.com,1999:blog-6141980.post-32564620523199768462010-12-13T12:08:18.149-06:002010-12-13T12:08:18.149-06:00Alejandro,
I would love to have somebody show me ...Alejandro,<br /><br />I would love to have somebody show me that they are indeed giving the same results. I am not being sarcastic here. <br /><br />Cheers,<br /><br />IgorIgorhttps://www.blogger.com/profile/17474880327699002140noreply@blogger.comtag:blogger.com,1999:blog-6141980.post-27145121448866255112010-12-13T11:13:13.465-06:002010-12-13T11:13:13.465-06:00"All l_1 recovery algorithms are not behaving..."All l_1 recovery algorithms are not behaving the same way. Few authors uses the proprietary code Mosek to get to these phase transitions."<br /><br />Why using Mosek may change the results? I thought that Mosek is just one (very fast) solver, but other than the execution time, it should produce the same results as CVX or any other solver.Alejandro Weinsteinnoreply@blogger.com